Just two months ago, Google’s DeepMind project shocked the tech world by announcing that its Go-playing artificial intelligence, AlphaGo, had successfully beaten a professional Go player, a feat that AI researchers had previously thought was still years away.

Now, AlphaGo has once again surprised the AI community by successfully beating the Go world champion, Lee Sedol, in its first of five matches.

“I was very surprised because I did not think that I would lose the game,” Sedol said after his first match against Google’s AI. “A mistake I made at the very beginning lasted until the very last.”

He added, “I had a lot of fun playing Go and I’m looking forward to the future games.”

AlphaGo’s success has major implications for the future of AI development, especially in the field of machine learning, which was how Google trained AlphaGo to be successful at the game. In a paper published in the journal Nature, the AlphaGo team explained that the AI had been fed data from 30 million moves from real Go games, after which it could predict human moves with 57 percent accuracy.

AlphaGo then played thousands of games against itself between its various neural networks, progressively strengthening its predictive abilities. Finally, Google pitted AlphaGo against a number of other Go-playing AI programs, and it won all but one of the 500 games it played.

“Lee Sedol and AlphaGo are almost at the same level”

Shortly after the results of AlphaGo’s first match against a professional player were made public in January, 9 dan ranked Go player Myungwan Kim analyzed the AI’s moves and said that while it was a very strong player, he did not believe that it would beat Sedol.

“No offense [to Google], but it’s not going to happen,” Kim said at the time.

Now however, Kim has admitted that AlphaGo has made major strides in strengthening its playstyle over the last few months, and he believes that the AI may be nearly equal to Sedol.

“Lee Sedol and AlphaGo are almost at the same level,” Kim said during a livestream of the recent match. “Even if [AlphaGo] is slightly weaker, it’s not by far.” When asked if he thought AlphaGo was slightly stronger or slightly weaker than Sedol, Kim joked, “From my analysis, one thing for sure is AlphaGo is stronger than me.”

Kim noted that AlphaGo did make several minor mistakes during its game, but none were catastrophic enough to make the AI lose.

AlphaGo’s next match against Lee Sedol will take place tonight at 9 p.m. PST.

Photo by chadmiller